Role of information technology in a study on antithrombotic-related bleeding events

With the emergence of computer-based patient records, it is to be expected that more retrospective studies will be conducted based on electronically stored clinical data. By using clinically collected data as research data, new opportunities and also new challenges emerge. The challenges include assessing the quality of the clinical data for research purposes and preparing the data to meet the needs of a study. The opportunities include finding novel ways to increase the power of a study based on the larger numbers of patients that can be identified retrospectively in active, enterprise-scale clinical databases. This dissertation demonstrates how information technology can be applied to address a subset of these challenges and opportunities in a study on antithrombotic-related bleeding events. A semi-automated process is described in which different computer programs prepared the data from the HELP system, the hospital information system at LDS Hospital, Salt Lake City, Utah, for a research database of patients taking antithrombotic medications. In 43.3% of patients data errors were detected and corrected through the semi-automated process. A novel type of electronic chart review is presented in which physician-reviewers used an electronic tool to classify the different types of bleeding events in a sample of patients. The tool presented the physician-reviewers with pertinent clinical data and allowed for the entry of their assessments. The tool ensured that the physician-reviewers followed a set of classification criteria for bleeding events. Nevertheless, it did not bring about the anticipated high level of agreement on the classification of bleeding events, because not all of the classification problems had been covered by the classification criteria. Finally, a classification program to identify the different types of bleeding events is presented. The program performed the same classification task as the physician-reviewers did. In a series of simulation experiments, I have demonstrated that it is feasible to conduct statistical analyses on data that had been classified by a program. Because the program could classify an entire study population, this approach allows for more power to detect small effects than a study restricted to a small sample that had been reviewed by physicians.